R：如何将“变量”的子集广播到data.table中的单独列？

Question

I have a big dataset in a data table that I am trying to transform. The original dataset is a crosstab with 1 to 2 levels of information in the column_names. So I thought I need to melt everything down, extract the relevant information, then re-cast the individual columns back.

This is where I am hitting a roadblock.

Below is a simplified mock data showing what I am trying to do:

Go from:

   ID1 ID2 X.Measure1 X.Measure2  Y.Measure1  Y.Measure2
1:   1   1 -0.6264538  0.4874291 -0.62124058  0.82122120
2:   2   2  0.1836433  0.7383247 -2.21469989  0.59390132
3:   3   1 -0.8356286  0.5757814  1.12493092  0.91897737
4:   1   2  1.5952808 -0.3053884 -0.04493361  0.78213630
5:   2   1  0.3295078  1.5117812 -0.01619026  0.07456498
6:   3   2 -0.8204684  0.3898432  0.94383621 -1.98935170

Perform 2 intermediate steps: (i) extract the integers '1' and '2' into a new column 'n'; and (ii) rename 'variable' to 'Y.Measure' (shown below on the left).

The final form is obtained by casting the figures in green as shown below on the right :

Sample Code:

library( data.table )
library( reshape2 )
library( stringr )

set.seed(1)
DT <- data.table( ID1 = rep( c(1:3),2 ), ID2 = rep( c(1:2),3 ), 
                  X.Measure1 = rnorm(6),  X.Measure2 = rnorm(6),
                  Y.Measure1 = rnorm(6),  Y.Measure2 = rnorm(6)
                  )

Long_DT <- melt( DT, id = c( "ID1", "ID2" ) )
Long_DT[ , n := substr( Long_DT$variable, 10, 10 ) ]

Long_DT[ str_detect( Long_DT$variable, "Y.Measure." ), variable := "Y.Measure"  ]

The Problem:

But when I tried dcast with a subset argument, I get the wrong result:

> dcast.data.table ( Long_DT, ID1+ID2 ~ variable, subset = (variable=="Y.Measure") )

Aggregate function missing, defaulting to 'length'
   ID1 ID2 Y.Measure
1:   1   1         2
2:   1   2         2
3:   2   1         2
4:   2   2         2
5:   3   1         2
6:   3   2         2

I tried Googling for the solution but to no avail. I am wondering if my dcast function is wrong or if my approach is wrong to begin with (ie there is a much easier way to achieve what I want).

Any help would be most appreciated! Thanks for reading!

---

UPDATE:

I found the error in my dcast function above - there should have been 'n' on the LHS:

dcast.data.table ( Long_DT, ID1+ID2+n ~ variable, subset = .(variable=="Y.Measure") )

The result would be:

> dcast.data.table ( Long_DT, ID1+ID2+n ~ variable, subset = .(variable=="Y.Measure") )
    ID1 ID2 n   Y.Measure
 1:   1   1 1 -0.62124058
 2:   1   1 2  0.82122120
 3:   1   2 1 -0.04493361
 4:   1   2 2  0.78213630
 5:   2   1 1 -0.01619026
 6:   2   1 2  0.07456498
 7:   2   2 1 -2.21469989
 8:   2   2 2  0.59390132
 9:   3   1 1  1.12493092
10:   3   1 2  0.91897737
11:   3   2 1  0.94383621
12:   3   2 2 -1.98935170
> 

Unfortunately, XMeasure1 and XMeasure2 also disappeared with the subset , so this doesn't help my overall cause.

Answer 1

Below is my modified code with akrun's suggested dcast code:

library( data.table )
library( reshape2 )
library( stringr )

set.seed(1)
DT <- data.table( ID1 = rep( c(1:3),2 ), ID2 = rep( c(1:2),3 ), 
                  X.Measure1 = rnorm(6),  X.Measure2 = rnorm(6),
                  Y.Measure1 = rnorm(6),  Y.Measure2 = rnorm(6)
                  )

Long_DT <- melt( DT, id = c( "ID1", "ID2" ) )
Long_DT[ , n := substr( Long_DT$variable, 10, 10 ) ]

Long_DT[ str_detect( Long_DT$variable, "Y.Measure." ), variable := "Y.Measure"  ]
dcast.data.table(Long_DT[, N:=1:.N, variable], ID1+ID2+N~variable, subset = (variable=="Y.Measure") )

Results:

    ID1 ID2  N   Y.Measure
 1:   1   1  1 -0.62124058
 2:   1   1  7  0.82122120
 3:   1   2  4 -0.04493361
 4:   1   2 10  0.78213630
 5:   2   1  5 -0.01619026
 6:   2   1 11  0.07456498
 7:   2   2  2 -2.21469989
 8:   2   2  8  0.59390132
 9:   3   1  3  1.12493092
10:   3   1  9  0.91897737
11:   3   2  6  0.94383621
12:   3   2 12 -1.98935170

Answer 2

I'm not sure if this is what you're expecting, but I just pushed a new feature to melt.data.table , that allows melting into multiple columns now..

You can install the development version by following these instructions . Then you can do:

require(data.table) ## v1.9.5
melt(DT, id=1:2, measure=list(3:4, 5:6), 
       value.name = c("X.measure", "Y.measure"))

By default, the variable column is populated with numbers. If that's not desirable, just change the levels of the variable column accordingly.

HTH